Image models understand the content of a prompt well and spatial relations badly. A request of the "product on the left, empty space on the right for the copy" kind regularly comes back the other way round, because the model does not process those words as positions. Layout control is the work with which you force it anyway.
Why does this matter commercially? Because composition is not decoration: an ad, a cover or a product shot needs space set aside in advance for the logo and the copy. If the composition cannot be steered, most of your images will be unusable, however pretty each one is on its own.
The most common misconception is that better prompting solves it. It does not — this is the weakest capability of today's image models, and the reliable answers are not in the prompt but next to it: a sketch, a reference composition, or laying the text on afterwards over a deliberately empty area.
The four workable methods are covered in layout control in images, which also says which one works almost every time. The directly related terms: ControlNet, reference image and character consistency.